Knowledge Mining Using Intelligent Agents

The following sections are included:

• PREFACE

• CONTENTS

Studying the behaviour of intelligent agents and deploy in various domain of interest with gamut of data to extract domain specific knowledge is recently attracting more and more number of researchers. In this chapter, we will summarize a few fundamental aspects of knowledge mining, the fundamental tasks of knowledge mining from databases (KMD) and a few well developed intelligent agents methodologies.

This chapter discusses the use of evolutionary computation in data mining and knowledge discovery by using intrusion detection systems as an example. The discussion centers around the role of EAs in achieving the two highlevel primary goals of data mining: prediction and description. In particular, classification and regression tasks for prediction, and clustering tasks for description. The use of EAs for feature selection in the pre-processing step is also discussed. Another goal of this chapter was to show how basic elements in EAs, such as representations, selection schemes, evolutionary operators, and fitness functions have to be adapted to extract accurate and useful patterns from data in different data mining tasks.

Natural evolution is the process of optimizing the characteristics and architecture of the living beings on earth. Possibly evolving the optimal characteristics and architectures of the living beings are the most complex problems being optimized on earth since time immemorial. The evolutionary technique though seems to be very slow, is one of the most powerful tools for optimization, especially when all the existing traditional techniques fail. This chapter presents how these evolutionary techniques can be used to generate optimal architecture and characteristics of different machine learning techniques. Mainly two different types of networks considered in this chapter for evolution are Artificial Neural Network and Polynomial Network. Though research has been conducted on evolution of Artificial Neural Network, research on evolution of polynomial Networks is still in its early stage. Evolution of both the networks have been discussed in detail. Experimental results are presented for further clarification of the evolution process of such networks.

In this chapter, a multi-objective optimization approach is used to address the alloy design problem, which concerns finding optimal processing parameters and the corresponding chemical compositions to achieve certain pre-defined mechanical properties of alloy steels. Neurofuzzy modeling has been used to establish the property prediction models for use in the multi-objective optimal design approach which is implemented using Particle Swarm Optimization (PSO). PSO is used as the search algorithm, because its populationbased approach fits well with the needs of multi-objective optimization. An evolutionary adaptive PSO algorithm is introduced to improve the performance of the standard PSO. Based on the established tensile strength and impact toughness prediction models, the proposed optimization algorithm has been successfully applied to the optimal design of heat-treated alloy steels. Experimental results show that the algorithm can locate the constrained optimal solutions quickly and provide a useful and effective guide for alloy steels design.

This chapter presents a hybrid adaptive particle swarm optimization (HAPSO)/Bayesian classifier to construct an intelligent and more compact intrusion detection system (IDS). An IDS plays a vital role of detecting various kinds of attacks in a computer system or network. The primary goal of the proposed method is to maximize detection accuracy with a simultaneous minimization of number attributes, which inherently reduces the complexity of the system. The proposed method can exhibits an improved capability to eliminate spurious features from huge amount of data aiding researchers in identifying those features that are solely responsible for achieving high detection accuracy. Experimental results demonstrate that the hybrid intelligent method can play a major role for detection of attacks intelligently.

Today networking of computing infrastructures across geographical boundaries has made it possible to perform various operations effectively irrespective of application domains. But, at the same time the growing misuse of this connectively in the form of network intrusions has jeopardized the security aspect of both the data that are transacted over the network and maintained in data stores. Research is in progress to detect such security threats and protect the data from misuse. A huge volume of data on intrusion is available which can be analyzed to understand different attack scenarios and devise appropriate counter-measures. The DARPA KDDcup'99 intrusion data set is a widely used data source which depicts many intrusion scenarios for analysis. This data set can be mined to acquire adequate knowledge about the nature of intrusions thereby one can develop strategies to deal with them. In this work we discuss on the use of different data mining techniques to elicit sufficient information that can be effectively used to build intrusion detection systems.

This chapter presents a particle swarm optimization for multi-objective optimal operational planning of energy plants. The optimal operational planning problem can be formulated as a mix-integer nonlinear optimization problem. An energy management system called FeTOP, which utilizes the presented method, is also introduced. FeTOP has been actually introduced and operated at three factories of one of the automobile companies in Japan and realized 10% energy reduction.

Feature selection has been the focus of interest for quite some time and much work has been done. It is in demand in areas of application for high dimensional datasets with tens or hundreds of thousands of variables available. This survey is a comprehensive overview of many existing methods from the 1970s to the present; considering both soft and non-soft computing paradigm. The strengths and weaknesses of different methods are explained and methods are categorized according to generation procedures and evaluation functions. The objective of feature selection is three fold: improving the prediction performance of the predictors, providing faster and more cost-effective prediction and providing a better understanding of the underlying process that generated the data. This survey identifies future research areas in feature subset selection and introduces newcomers to this field.

This chapter presents a hybrid approach for solving classification problems. We have used three important neuro and evolutionary computing techniques such as Polynomial Neural Network, Fuzzy system, and Particle Swarm Optimization to design a classifier. The objective of designing such a classifier model is to overcome some of the drawbacks in the existing systems and to obtain a model that consumes less time in developing the classifier model, to give better classification accuracy, to select the optimal set of features required for designing the classifier and to discard less important and redundant features from consideration. Over and above the model remains comprehensive and easy to understand by the users.

Traditional software testing methods involve large amounts of manual tasks which are expensive in nature. Software testing effort can be significantly reduced by automating the testing process. A key component in any automatic software testing environment is the test data generator. As test data generation is treated as an optimization problem, Genetic Algorithm has been used successfully to generate automatically an optimal set of test cases for the software under test. This chapter describes a framework that automatically generates an optimal set of test cases to achieve path coverage of an arbitrary program.